What I Expect From an AI Assistant in a Management Product

I build JMP on my own. It works with operational data, and I added a conversational assistant to it because I kept seeing the same problem: getting from a question to a useful decision takes too many steps.

Someone may see a change in one metric and want to understand it. The answer is often not in that one view. They need to open other reports, check how the numbers are calculated, compare periods, and then try to find a pattern. Only after this work can they form a hypothesis about what is happening.

The assistant can help with the preparation. A person can ask a question in plain language. The system can get the relevant data, do a calculation, compare values, and point to something that deserves attention.

This is useful, but I do not see it as an autonomous management system.

An answer is not a decision

It is easy to make a chat interface look convincing. It can produce a clean explanation of what happened in the data. It may even suggest a few next steps. But a management decision needs more than a good explanation.

The data does not know what the team has already promised. It does not know who is overloaded, what depends on a planned change, or how expensive a wrong decision may be. It cannot decide which trade-off is acceptable in a specific situation.

These details are not outside the real work. They are the real work.

For example, an assistant may show that a certain type of work is taking longer than before. This is a useful signal. But there can be many reasons: a team may be working on a harder part of the product, handling an incident, helping another team, or changing its process. The numbers can start the conversation. They cannot finish it.

The part worth automating

I think there is a clear role for AI here. It should reduce the mechanical work that happens before judgment.

That includes finding data, checking calculations, comparing reports, and bringing related signals together. These tasks are important, but they should not take most of a manager's attention. A manager's attention is more valuable when it is spent on context, priorities, and action.

This is also how I evaluate an AI feature. I do not start with the question: can it produce an answer that sounds intelligent? I start with a simpler question: does it help a person reach the important question faster?

Sometimes the right outcome is not a recommendation. It can be a better question, a pattern to investigate, or a gap in the available data. That is still useful because it makes the next conversation more concrete.

Keeping responsibility in the right place

There is a temptation to describe AI as something that will make decisions instead of people. I do not find this framing helpful for management products.

Managers are not only processing information. They are responsible for people, commitments, risks, and consequences. A system can support this work, but it should not hide that responsibility behind a confident answer.

The product I want to build does not try to remove the person from the loop. It tries to remove unnecessary distance between a question and the evidence needed to discuss it.

This boundary makes the assistant more honest. It also makes it more useful in daily work.

Anvar Khakimov — Product and engineering leader, 15+ years from developer to IT lead. Building JMP and writing about technology, management, and AI.